DEM landform analysis method based on large language model

Abstract

The invention discloses a DEM landform analysis method based on a large language model, and relates to the technical field of DEM data analysis, and the method comprises the steps: loading a DEM semantic mapping tool set, and achieving system initialization; after the system receives user input, questions are divided into single-tool questions, multi-tool collaborative questions and non-tool answerable questions through a question classification Agent on the basis of a pre-training model; respectively processing a single tool question, a multi-tool collaborative question and a non-tool answerable question; and performing existing memory retrieval on each problem based on a bidirectional knowledge evolution memory iteration mechanism, and performing multi-dimensional analysis in combination with the large language model to realize DEM landform analysis based on the large language model. The technical problems that in the prior art, LLM is difficult to directly and effectively analyze the spatial topological relation contained in DEM raster data, and iteration updating of geographic knowledge in a complex dynamic analysis task is difficult are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of DEM data analysis, and in particular to a DEM landform analysis method based on a large language model. Background Art

[0002] As a core component of geospatial information, digital elevation model (DEM) plays an irreplaceable fundamental role in key areas such as geological disaster monitoring and early warning, national land space and ecological environment planning, and provides key data support for many decision support systems.

[0003] However, existing geographic information system (GIS) tools generally rely on an expert experience-based analysis paradigm for DEM data analysis and application. This paradigm often exposes bottlenecks such as low efficiency and low automation when handling advanced tasks such as the fusion analysis of increasingly complex multimodal geospatial data and intelligent reasoning of complex terrain features.

[0004] In recent years, large language models (LLMs), leveraging significant advances in natural language processing and understanding unstructured data, have begun to demonstrate their potential for application in geospatial data analysis. Despite this, LLMs currently face a significant semantic gap when directly processing and understanding discrete raster data such as DEMs. LLMs inherently struggle to directly interpret the spatial topological characteristics inherent in raster data, such as spatial location, distribution, and relationships. This limits their effective application in deep terrain understanding and analysis tasks.

[0005] Against this backdrop, academia and industry have begun exploring technical approaches to integrating GIS and LLM, hoping to combine the strengths of both. However, existing fusion technology solutions, particularly in the in-depth analysis and application of DEM data, face the following core technical bottlenecks:

[0006] (1) Insufficient analysis and semantic understanding of DEM spatial topological relationships;

[0007] (2) There is a lack of a mechanism for dynamic iteration and efficient retrieval of geographic knowledge.

[0008] Therefore, how to effectively bridge the gap between the spatial characteristics of DEM raster data and the semantic understanding capabilities of LLM, and establish a knowledge iteration and efficient retrieval mechanism that adapts to the needs of dynamic geographic analysis, is a key technical problem that needs to be urgently solved in the current field of geographic spatial intelligence. Summary of the Invention

[0009] In view of the above deficiencies in the prior art, a DEM geomorphological analysis method based on a large language model provided by the present invention solves the technical problems in the prior art that it is difficult for an LLM to directly and effectively analyze the spatial topological relationships contained in DEM raster data, and it is difficult to iteratively update geographical knowledge in complex dynamic analysis tasks.

[0010] To achieve the above invention purpose, the technical solution adopted by the present invention is: a DEM geomorphological analysis method based on a large language model, including the following steps:

[0011] S1: Load the DEM semantic mapping tool set to realize system initialization;

[0012] S2: When the system receives user input, the question classification Agent classifies the question into a single tool question, a multi-tool collaboration question, and a non-tool answerable question based on a pre-trained large language model;

[0013] S3: Process the single tool question, the multi-tool collaboration question, and the non-tool answerable question respectively;

[0014] S4: Retrieve the existing memory of each question based on the bidirectional knowledge evolution memory iteration mechanism, and perform multi-dimensional analysis in combination with the large language model to realize DEM geomorphological analysis based on the large language model.

[0015] Further, the following sub-steps are included in S1:

[0016] S11: Load the DEM semantic mapping tool set, and the tool set is implemented based on the GDAL library;

[0017] S12: Use the tools in the DEM semantic mapping tool set to parse the DEM raster data through Python functions, and output a standardized semantic structure;

[0018] S13: Load the memory knowledge base, and the memory knowledge base includes a permanent memory base and a short-term memory base. During initialization, the timeliness evaluation module calculates the timeliness score of the short-term memory entries according to the time stamp, and performs memory forgetting according to the timeliness score to eliminate expired knowledge.

[0019] Further, the calculation formula of the timeliness score is:

[0020]

[0021] Among them, is the timeliness score, is the current time, is the memory time stamp, is the hyperparameter.

[0022] Further, the processing of single-tool problems in S3 includes the following sub-steps:

[0023] a1: The problem classification Agent parses the question raised by the user into a question-tool-parameter pair;

[0024] a2: The function call node executes the tool according to the parsing result, integrates the result into a natural language answer after obtaining it. If the tool call fails, the system returns an error message and suggests that the user adjust the input.

[0025] Further, the processing of multi-tool collaboration problems in S3 includes the following sub-steps:

[0026] b1: The problem classification Agent decomposes the question raised by the user based on the chain of thought method to generate a problem decomposition structure;

[0027] b2: Design a parallel tool call node, use multi-threading to parallelly call tools to call the problem decomposition structure, and return the tool call results to the large language model;

[0028] b3: The large language model combines the tool call results and the intermediate product path generated during the tool call to generate an output in markdown format.

[0029] Further, the processing method for non-tool-answerable questions in S3 is as follows: The system processes the question raised by the user based on the RAG paradigm of dual retrieval of permanent memory and short-term memory, specifically including:

[0030] c1: Retrieve the question raised by the user in the permanent memory library and the short-term memory library;

[0031] c2: Input the retrieval results into the large language model, and the large language model integrates the two types of retrieval results to generate an accurate answer.

[0032] Further, S4 includes the following sub-steps:

[0033] S41: The system retrieves the existing memory according to a specific question and conducts multi-dimensional analysis in combination with the large language model;

[0034] S42: The large language model scores the input memory entries and outputs a structure containing accuracy score, relevance score, integrity score, security score, whether it contains the subject, and repeatability score;

[0035] S43: Based on the structure, obtain the evaluation score through weighted formula fusion;

[0036] S44: Use the evaluation score, whether it contains the subject, and the repeatability score as the decision basis for memory update to perform short-term memory update;

[0037] S45: The system improves the timeliness of the next retrieval according to the updated memory, achieving two-way timeliness optimization.

[0038] Further, the calculation formula for the evaluation score in S43 is:

[0039]

[0040] Among them, is the evaluation score, and are the weight coefficients, is the timeliness correlation coefficient of the current memory, is the accuracy score, is the relevance score, is the integrity score, is the security score.

[0041] Further, the short-term memory update in S44 includes the following sub-steps:

[0042] S441: If the evaluation score of the current document satisfies ≥7 and repetition < 2.5 or subject == False, then go to step S442, where repetition represents the repetition score, subject represents whether the subject is included, and subject == False means that if the subject is not included in the current memory library, memory update is performed;

[0043] S442: The language large model automatically generates a role portrait of the agent according to the document content and determines its field or professional direction;

[0044] S443: According to the role portrait, hierarchically decompose the knowledge of the document, remove redundant content, refine the key points, and form a list of memory entries response;

[0045] S444: For each entry in the memory entry list response, construct the page content page_content in the format of "title:" + item ['subject'] + "-" + item ['title'] + "content:" + item ['content'], create a document doc with the page content as page_content, add the current memory entry timestamp to the metadata, store the document doc in the Chroma vector database, and use it as the short-term memory library to complete the short-term memory update.

[0046] The beneficial effects of the present invention are:

[0047] First, the present invention designs and implements a set of grid data terrain feature extraction and semanticization processes. Through structured module calls and processing, this process converts discrete DEM grid data into feature information with clear geographical semantics, including statistical features, terrain parameter fields, etc., thus effectively bridging the gap between the original DEM data and the semantic understanding of the LLM, and realizing the efficient mapping of geospatial data to the semantic space processable by the LLM.

[0048] Second, the present invention constructs a two-way knowledge evolution memory iteration mechanism. This mechanism innovatively integrates multi-dimensional consistency evaluation methods and knowledge hierarchy decomposition strategies, and can dynamically evaluate, screen, correct, and enhance relevant geographical knowledge in the internal or external knowledge base of the LLM according to the feedback and intermediate results of the analysis task, so as to realize the continuous optimization and adaptive evolution of knowledge during the analysis process, and improve the analysis ability for dynamic scenarios such as complex terrain evolution.

[0049] Through the synergistic effect of the above-mentioned grid data terrain feature extraction and semanticization process and the two-way knowledge evolution and memory iteration mechanism, the framework proposed by the present invention enables the large language model to more accurately and efficiently conduct in-depth analysis and understanding of the digital elevation model, and can automatically generate a structured and content-rich geomorphic analysis report based on this, significantly improving the level and application efficiency of intelligent analysis of DEM data. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a flowchart of a DEM geomorphic analysis method based on a large language model.

[0051] Figure 2 It is a flowchart for solving a single tool problem.

[0052] Figure 3 It is a schematic diagram of a multi-tool problem-solving and two-way memory optimization and update mechanism.

[0053] Figure 4 It is a flowchart for solving non-tool solvable problems. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] The present invention will be further described below with reference to the drawings and specific embodiments.

[0055] As Figure 1 shown, a DEM geomorphic analysis method based on a large language model includes the following steps:

[0056] S1: Load the DEM semantic mapping tool set to realize system initialization;

[0057] S2: After the system receives user input, the question classification Agent classifies the question into a single-tool question, a multi-tool collaboration question, and a non-tool answerable question based on a pre-trained large language model.

[0058] S3: Process single-tool questions, multi-tool collaboration questions, and non-tool answerable questions separately.

[0059] S4: Retrieve existing memories for each question based on the two-way knowledge evolution memory iteration mechanism, and perform multi-dimensional analysis in combination with the large language model to achieve DEM landform analysis based on the large language model.

[0060] The S1 includes the following sub-steps:

[0061] S11: Load the DEM semantic mapping tool set, which is implemented based on the GDAL library.

[0062] Specifically, it includes the following core tools, as shown in Table 1:

[0063] Table 1 Core Tool List

[0064] Tool Name Function Description Parameters Return Value open_dataset Open the DEM map according to the DEM path and load the DEM data dem_path: str (DEM file path) Map opening success information and basic information get_elevation Get the elevation information of a certain point x: float (longitude), y: float (latitude) Elevation information of the point or error prompt get_area_information Get the slope, elevation or aspect statistical information of the area min_lon: float, min_lat: float, max_lon: float, max_lat: float, task: str (task type) Area statistical information or error prompt get_slope_percentage Get the proportion of a specific slope range in the DEM min_slope: float, max_slope: float Slope proportion information or error prompt get_slope_distribution Get the slope distribution of the DEM None Slope distribution information plot_2d_elevation Visualize the digital elevation model and save it as a picture None Visualization image save path or error prompt get_geo_types Get the landform types according to the opened DEM None Landform type statistical information water_area_analysis Get the water area analysis according to the opened DEM and save it as a picture None Water area analysis results and image save path get_countours Get the landform contour analysis according to the opened DEM and save it as a picture None Contour map save path and statistical information

[0065] In Table 1, min_lon refers to the minimum longitude, with a floating-point type; min_lat refers to the minimum latitude, with a floating-point type; max_lon refers to the maximum longitude, with a floating-point type; max_lat refers to the maximum latitude, with a floating-point type; min_slope refers to the minimum value of the slope range, with a floating-point type; max_slope refers to the maximum value of the slope range, with a floating-point type.

[0066] S12: Use the tools in the DEM semantic mapping tool set to parse the DEM raster data through Python functions and output a standardized semantic structure.

[0067] In this embodiment, taking the slope statistics result as an example:

[0068] {

[0069] "Slope statistics information": {

[0070] "Maximum slope": "85 degrees",

[0071] "Minimum slope": "5 degrees",

[0072] "Average slope": "20.6 degrees",

[0073] "Slope standard deviation": "0.23",

[0074] "Slope Distribution": {"0-10": "30%", "10-20": "40%", "20-30": "10%", "30-50":"10%", "50-70": "6%", "70-90": "4%"}

[0075] }

[0076] }

[0077] S13: Load the memory knowledge base. The memory knowledge base includes a permanent memory base and a short-term memory base. During initialization, the timeliness evaluation module calculates the timeliness score of the short-term memory entries according to the timestamp, and performs memory forgetting according to the timeliness score to eliminate expired knowledge, ensuring that the final analysis is based on the latest data.

[0078] Among them, the permanent memory base stores objective geographical knowledge (such as landform classification rules). The short-term memory base stores timeliness knowledge (such as recent terrain change data).

[0079] The calculation formula for the timeliness score is:

[0080]

[0081] Among them, is the timeliness score, is the current time, is the memory timestamp, is a hyperparameter that determines the frequency of memory forgetting.

[0082] The memory forgetting logic is directly related to timeliness. The specific logic process is as follows:

[0083] First, retrieve all document IDs from the short-term memory base (kno_docsearch) to generate a list of original IDs, raw_ids;

[0084] Secondly, create an empty list valid_ids to store the IDs that meet the retention conditions;

[0085] Then, for each ID in raw_ids, perform the following judgment:

[0086] Call the should_keep_id function to determine whether the ID should be retained according to the time limit of max_interval days;

[0087] If the ID should be retained, add it to the valid_ids list.

[0088] Finally, if the length of the valid_ids list is greater than 0, delete all the IDs in the valid_ids list from the knowledge base document (kno_docsearch).

[0089] After the system receives the user input in S2, the question classification Agent classifies the question into the following categories based on the pre-trained model:

[0090] Single-tool question: such as "Calculate the slope distribution in a certain area".

[0091] Multi-tool collaborative question: such as "Summarize the geomorphic features of a certain area".

[0092] Non-tool-answerable question: such as "Explain the cause of a certain topographic feature".

[0093] The classification result guides the subsequent processing path and improves the system efficiency.

[0094] Among them, for a single-tool question, that is, a single tool can be used to complete the user's question;

[0095] For a multi-tool collaborative question, it means that the user's question needs to be executed by multiple tools in collaboration to complete;

[0096] For a non-tool-answerable question, it means that the user's question can be answered without using a tool, relying on their own ability or the knowledge in the memory library.

[0097] Specifically, when the user inputs a question to the system, the large model will analyze the user's question and output according to the following rules: for a single-tool question, output {"problem_type":1}; for a multi-tool collaborative question, output {"problem_type":2}; for a non-tool-answerable question, output {"problem_type":3}. The system will subsequently judge the type of the current user's question based on problem_type.

[0098] Such as Figure 2 As shown, the processing of a single-tool question in S3 includes the following sub-steps:

[0099] a1: The question classification Agent parses the question raised by the user into a question-tool-parameter pair;

[0100] Taking the calculation of slope distribution as an example, its format is as follows:

[0101] {

[0102] "task": "Calculate slope distribution",

[0103] "action": "get_slope_distribution",

[0104] "tool_param": null

[0105] }

[0106] a2: The function call node executes the tool according to the parsing result, and after obtaining the result, it integrates it into a natural language answer. If the tool call fails, the system returns an error message and suggests that the user adjust the input.

[0107] Such as Figure 3 shown, the processing of the multi-tool collaboration problem in S3 includes the following sub-steps:

[0108] b1: The problem classification Agent decomposes the question raised by the user based on the chain of thought method to generate a problem decomposition structure;

[0109] Taking the regional landslide risk as an example, the generated problem decomposition structure is as follows:

[0110] {

[0111] "Task": "Evaluate the landslide risk in a certain area",

[0112] "Steps":

[0113] {"Subtask": "Extract elevation data", "Action": "get_elevation", "Tool_param":{"min_lon": 120.1, "max_lat": 30.2, ...}},

[0114] {"Subtask": "Calculate slope distribution", "Action": "get_slope_distribution", "Tool_param": null},

[0115] {"Subtask": "Analyze the influence of water area", "Action": "water_area_analysis", "Tool_param": null},

[0116] …,

[0118] }

[0119] Among them, Task is the task proposed by the user, and Steps is a list containing multiple subtasks. Each subtask has attributes such as subtask description, the function tool to be taken, and tool parameters.

[0120] b2: Design a parallel tool call node, use multi-threading to parallelly call the tools to call the problem decomposition structure, and return the tool call results to the large language model;​

[0121] During tool calls, a series of intermediate products will be generated, such as elevation maps, slope aspect maps, water area analysis maps, contour maps, etc. These products will be named with unique codes and returned to the large model through the tool call results.

[0122] In this embodiment, a parallel tool call node is designed, and the specific process is as follows:

[0123] (1) Sub-task status preparation: Receive the sub-task list of the current task. For each sub-task in the list, the module creates an independent status copy dedicated to processing this sub-task. This independent status contains information about the sub-task as the current input and processing context.

[0124] (2) Instantiate the sub-agent workflow: The node defines and instantiates a simplified sub-agent workflow. This sub-workflow is designed to process a single sub-task. The node is responsible for calling one or more external tools according to the specific content of the sub-task and obtaining the tool execution results.

[0125] (3) Parallel execution: Using a parallel execution mechanism (e.g., ThreadPoolExecutor), the module assigns an execution thread or process to each prepared sub-task status and its corresponding sub-agent workflow instance. All these sub-agent workflow instances are started and executed concurrently. Each instance runs in its independent context, calling the tool call nodes it contains to process the sub-tasks assigned to it.

[0126] (4) Result collection and aggregation: The node waits for all parallelly executed sub-agent workflows to complete. Collect the execution results of each completed instance, especially the output information obtained from its tool call nodes. For sub-tasks that execute successfully and return valid results, their results are retained; for sub-tasks that execute fails or do not return valid results, they can be marked or filtered out. Finally, the results of all successfully executed sub-tasks are aggregated into a list and updated to the original agent status for subsequent processes (such as result synthesis, final answer generation, etc.) to use.

[0127] b3: The large language model combines the tool call results and the paths of the intermediate products generated during tool calls to generate an output in markdown format.

[0128] As Figure 4 shown, the processing method for questions that cannot be answered by tools in S3 is as follows: The system processes the questions raised by users based on the RAG paradigm of dual retrieval of permanent memory (objective truth) and short-term memory (timely information), specifically including:

[0129] c1: Retrieve the questions raised by the user in the permanent memory bank and the short-term memory bank;

[0130] c2: Input the retrieval results into the large language model, and the large language model integrates the two types of retrieval results to generate accurate answers.

[0131] For example, when the user asks "the reasons for the recent topographic changes in a certain area", the system provides a comprehensive answer by combining permanent knowledge (the principle of landform formation) and short-term memory (recent monitoring data). The short-term memory will be updated in a timely manner to ensure the timeliness of the answer.

[0132] The answers generated by the multi-tool question are obtained from the DEM elevation analysis tool and are highly credible. The short-term memory can be effectively managed and updated through reverse optimization technology.

[0133] In the two-way optimization mechanism, the system can retrieve existing memories according to specific questions and conduct multi-dimensional analysis in combination with the large language model to ensure the requirements in terms of accuracy, relevance, integrity, security, subject coverage, repeatability, and timeliness of the results. The specific operation process is as follows:

[0134] The S4 includes the following sub-steps:

[0135] S41: The system retrieves existing memories according to specific questions and conducts multi-dimensional analysis in combination with the large language model;

[0136] S42: The large language model scores the input memory entries and outputs a structure containing accuracy scores, relevance scores, integrity scores, security scores, whether the subject is included, and repeatability scores:

[0137] {

[0138] "accuracy": <type:float>, / / Accuracy score (1-10 points)

[0139] "relevance": <type:float>, / / Relevance score (1 - 10 points)

[0140] "completeness": <type:float>, / / Integrity score (1 - 10 points)

[0141] "safety": <type:float>, / / Security score (1 - 10 points)

[0142] "subject": <type:bool>, / / Whether the subject is included (boolean value)

[0143] "repetition": <type:float> / / Repeatability score (1-10 points)

[0144] }

[0145] S43: Based on the structure, the evaluation score is obtained by fusion of weighted formulas;

[0146] The calculation formula of the evaluation score in S43 is:

[0147]

[0148] in, To evaluate the score, it is used to balance timeliness and multi-dimensional scoring. and is the weight coefficient, is the timeliness correlation coefficient of current memory, Score for accuracy, is the correlation score, Score for completeness, Score for safety;

[0149] S44: Using the assessment score, whether to include the subject, and the repeatability score as the basis for memory updating decisions, short-term memory updating is performed;

[0150] Before updating short-term memory, the agent will simulate the human knowledge comprehension model, first performing a hierarchical decomposition of knowledge, removing redundant content, extracting the core points, forming knowledge point entries, and finally using the knowledge point entries to update short-term memory. The specific logic of short-term memory update is as follows:

[0151] The short-term memory updating in S44 includes the following steps:

[0152] S441: If the evaluation score of the current document satisfies ≥7 and repetition<2.5 or subject==False, then proceed to step S442, where repetition represents the repeatability score, subject represents whether the subject is included, and subject==False means that if the current memory library does not contain the subject, then the memory is updated;

[0153] S442: The language model automatically generates a role profile for the agent based on the document content, determining its field or professional direction (e.g., terrain classification expert, water area analysis expert, etc.);

[0154] S443: Based on the role profile, decompose the document into a knowledge hierarchy, remove redundant content, extract key points, and form a memory item list response;

[0155] S444: For each entry in the memory entry list response, construct the page content page_content in the format of "title:" + item['subject'] + "-" + item['title'] + "content:" + item['content'], create a document doc with the page content as page_content, add the timestamp of the current memory entry to the metadata, store the document doc in the Chroma vector database as a short-term memory bank, and complete the short-term memory update;

[0156] S45: After the short-term memory update is completed, the system improves the timeliness of the next retrieval based on the updated memory, realizes two-way timeliness optimization, and improves the response accuracy and reliability of the system.

[0157] Those of ordinary skill in the art will realize that the embodiments described herein are for helping readers understand the principles of the present invention and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those of ordinary skill in the art can make various other specific deformations and combinations without departing from the essence of the present invention based on the technical revelations disclosed in the present invention, and these deformations and combinations are still within the protection scope of the invention.< / type:float> < / type:bool> < / type:float> < / type:float> < / type:float> < / type:float>

Claims

1. A DEM landform analysis method based on large language models, characterized in that, It includes the following steps: S1: Load the DEM semantic mapping toolset to initialize the system; S2: When the system receives user input, the question classification Agent classifies the question into single-tool questions, multi-tool collaborative questions, and non-tool solvable questions based on a pre-trained large language model; S3: Process single-tool questions, multi-tool collaborative questions, and non-tool solvable questions separately; S4: Retrieve existing memories for each question based on the bidirectional knowledge evolution memory iteration mechanism, and perform multi-dimensional analysis in combination with the large language model to achieve DEM landform analysis based on the large language model.

2. The DEM landform analysis method based on a large language model according to claim 1, characterized in that, The following sub-steps are included in S1: S11: Load the DEM semantic mapping toolset, which is implemented based on the GDAL library; S12: Use the tools in the DEM semantic mapping toolset to parse the DEM raster data through Python functions and output a standardized semantic structure; S13: Load the memory knowledge base, which includes a permanent memory base and a short-term memory base. During initialization, the timeliness evaluation module calculates the timeliness score of short-term memory entries according to the timestamp, and performs memory forgetting according to the timeliness score to eliminate expired knowledge.

3. The DEM landform analysis method based on a large language model according to claim 2, wherein The calculation formula for the timeliness score is: wherein, is the timeliness score, is the current time, is the memory timestamp, is a hyperparameter.

4. The DEM geomorphic analysis method based on the large language model according to claim 1, wherein The processing of single-tool questions in S3 includes the following sub-steps: a1: The question classification Agent parses the question raised by the user into a question-tool-parameter pair; a2: The function call node executes the tool according to the parsing result, integrates the results into a natural language answer after obtaining the results. If the tool call fails, the system returns an error message and suggests that the user adjust the input.

5. The DEM geomorphic analysis method based on the large language model according to claim 1, characterized in that, The processing of multi-tool collaborative questions in S3 includes the following sub-steps: b1: The question classification Agent decomposes the question raised by the user based on the chain of thought method to generate a question decomposition structure; b2: Design a parallel tool call node, use multi-threading to parallelly call the tool to call the question decomposition structure, and return the tool call results to the large language model; b3: The large language model combines the tool call results and the intermediate product path generated during the tool call to generate an output in markdown format.

6. The DEM landform analysis method based on a large language model according to claim 1, wherein, The processing method for non-tool solvable questions in S3 is: The system processes the question raised by the user based on the RAG paradigm of dual retrieval of permanent memory and short-term memory, specifically including: c1: Retrieve the question raised by the user in the permanent memory base and the short-term memory base; c2: Input the retrieval results into the large language model, and the large language model integrates the two types of retrieval results to generate an accurate answer.

7. The DEM landform analysis method based on the large language model according to claim 1, wherein The following sub-steps are included in S4: S41: The system retrieves existing memories according to specific questions and performs multi-dimensional analysis in combination with the language large model; S42: The language large model scores the input memory entries and outputs a structure including accuracy score, relevance score, integrity score, security score, whether it contains the subject, and repetition score; S43: Based on the structure, fuse and obtain an evaluation score through a weighted formula S44: Use the evaluation score, whether the subject is included, and the repeatability score as the decision basis for memory update to perform short-term memory update; S45: The system improves the timeliness of the next retrieval based on the updated memory, achieving two-way timeliness optimization.

8. The DEM landform analysis method based on a large language model according to claim 7, wherein The calculation formula of the evaluation score in S43 is: Among them, is the evaluation score, and are the weight coefficients, is the timeliness correlation coefficient of the current memory, is the accuracy score, is the relevance score, is the integrity score, is the security score.

9. The DEM landform analysis method based on a large language model according to claim 8, wherein, The short-term memory update in S44 includes the following sub-steps: S441: If the evaluation score of the current document meets ≥ 7 and repetition < 2.5 or subject == False, then proceed to step S442, where repetition represents the repetition score, subject represents whether the subject is included, and subject == False means that if the subject is not included in the current memory bank, memory update is performed; S442: The language large model automatically generates a role portrait of the agent based on the document content to determine its field or professional direction; S443: According to the role portrait, hierarchically decompose the knowledge of the document, remove redundant content, extract key points, and form a list of memory entries response; S444: For each entry in the memory entry list response, construct the page content page_content in the format of "title:"+item ['subject']+"-"+item ['title']+"content:"+item ['content'], and create a document doc with the page content as page_content. Add the timestamp of the current memory entry to the metadata, store the document doc in the Chroma vector database, and use it as the short-term memory library to complete the short-term memory update.

Images